drivers/scsi/scsi_error.c

   1 /*
   2  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
   3  *
   4  *  SCSI error/timeout handling
   5  *      Initial versions: Eric Youngdale.  Based upon conversations with
   6  *                        Leonard Zubkoff and David Miller at Linux Expo,
   7  *                        ideas originating from all over the place.
   8  *
   9  *      Restructured scsi_unjam_host and associated functions.
  10  *      September 04, 2002 Mike Anderson (andmike@us.ibm.com)
  11  *
  12  *      Forward port of Russell King's (rmk@arm.linux.org.uk) changes and
  13  *      minor  cleanups.
  14  *      September 30, 2002 Mike Anderson (andmike@us.ibm.com)
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/sched.h>
  19 #include <linux/gfp.h>
  20 #include <linux/timer.h>
  21 #include <linux/string.h>
  22 #include <linux/kernel.h>
  23 #include <linux/freezer.h>
  24 #include <linux/kthread.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/delay.h>
  28
  29 #include <scsi/scsi.h>
  30 #include <scsi/scsi_cmnd.h>
  31 #include <scsi/scsi_dbg.h>
  32 #include <scsi/scsi_device.h>
  33 #include <scsi/scsi_eh.h>
  34 #include <scsi/scsi_transport.h>
  35 #include <scsi/scsi_host.h>
  36 #include <scsi/scsi_ioctl.h>
  37
  38 #include "scsi_priv.h"
  39 #include "scsi_logging.h"
  40 #include "scsi_transport_api.h"
  41
  42 #include <trace/events/scsi.h>
  43
  44 #define SENSE_TIMEOUT           (10*HZ)
  45
  46 /*
  47  * These should *probably* be handled by the host itself.
  48  * Since it is allowed to sleep, it probably should.
  49  */
  50 #define BUS_RESET_SETTLE_TIME   (10)
  51 #define HOST_RESET_SETTLE_TIME  (10)
  52
  53 /* called with shost->host_lock held */
  54 void scsi_eh_wakeup(struct Scsi_Host *shost)
  55 {
  56         if (shost->host_busy == shost->host_failed) {
  57                 trace_scsi_eh_wakeup(shost);
  58                 wake_up_process(shost->ehandler);
  59                 SCSI_LOG_ERROR_RECOVERY(5,
  60                                 printk("Waking error handler thread\n"));
  61         }
  62 }
  63
  64 /**
  65  * scsi_schedule_eh - schedule EH for SCSI host
  66  * @shost:      SCSI host to invoke error handling on.
  67  *
  68  * Schedule SCSI EH without scmd.
  69  */
  70 void scsi_schedule_eh(struct Scsi_Host *shost)
  71 {
  72         unsigned long flags;
  73
  74         spin_lock_irqsave(shost->host_lock, flags);
  75
  76         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
  77             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
  78                 shost->host_eh_scheduled++;
  79                 scsi_eh_wakeup(shost);
  80         }
  81
  82         spin_unlock_irqrestore(shost->host_lock, flags);
  83 }
  84 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
  85
  86 /**
  87  * scsi_eh_scmd_add - add scsi cmd to error handling.
  88  * @scmd:       scmd to run eh on.
  89  * @eh_flag:    optional SCSI_EH flag.
  90  *
  91  * Return value:
  92  *      0 on failure.
  93  */
  94 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
  95 {
  96         struct Scsi_Host *shost = scmd->device->host;
  97         unsigned long flags;
  98         int ret = 0;
  99
 100         if (!shost->ehandler)
 101                 return 0;
 102
 103         spin_lock_irqsave(shost->host_lock, flags);
 104         if (scsi_host_set_state(shost, SHOST_RECOVERY))
 105                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
 106                         goto out_unlock;
 107
 108         ret = 1;
 109         scmd->eh_eflags |= eh_flag;
 110         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
 111         shost->host_failed++;
 112         scsi_eh_wakeup(shost);
 113  out_unlock:
 114         spin_unlock_irqrestore(shost->host_lock, flags);
 115         return ret;
 116 }
 117
 118 /**
 119  * scsi_times_out - Timeout function for normal scsi commands.
 120  * @req:        request that is timing out.
 121  *
 122  * Notes:
 123  *     We do not need to lock this.  There is the potential for a race
 124  *     only in that the normal completion handling might run, but if the
 125  *     normal completion function determines that the timer has already
 126  *     fired, then it mustn't do anything.
 127  */
 128 enum blk_eh_timer_return scsi_times_out(struct request *req)
 129 {
 130         struct scsi_cmnd *scmd = req->special;
 131         enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
 132
 133         trace_scsi_dispatch_cmd_timeout(scmd);
 134         scsi_log_completion(scmd, TIMEOUT_ERROR);
 135
 136         if (scmd->device->host->transportt->eh_timed_out)
 137                 rtn = scmd->device->host->transportt->eh_timed_out(scmd);
 138         else if (scmd->device->host->hostt->eh_timed_out)
 139                 rtn = scmd->device->host->hostt->eh_timed_out(scmd);
 140
 141         if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
 142                      !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
 143                 scmd->result |= DID_TIME_OUT << 16;
 144                 rtn = BLK_EH_HANDLED;
 145         }
 146
 147         return rtn;
 148 }
 149
 150 /**
 151  * scsi_block_when_processing_errors - Prevent cmds from being queued.
 152  * @sdev:       Device on which we are performing recovery.
 153  *
 154  * Description:
 155  *     We block until the host is out of error recovery, and then check to
 156  *     see whether the host or the device is offline.
 157  *
 158  * Return value:
 159  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
 160  */
 161 int scsi_block_when_processing_errors(struct scsi_device *sdev)
 162 {
 163         int online;
 164
 165         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
 166
 167         online = scsi_device_online(sdev);
 168
 169         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
 170                                           online));
 171
 172         return online;
 173 }
 174 EXPORT_SYMBOL(scsi_block_when_processing_errors);
 175
 176 #ifdef CONFIG_SCSI_LOGGING
 177 /**
 178  * scsi_eh_prt_fail_stats - Log info on failures.
 179  * @shost:      scsi host being recovered.
 180  * @work_q:     Queue of scsi cmds to process.
 181  */
 182 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
 183                                           struct list_head *work_q)
 184 {
 185         struct scsi_cmnd *scmd;
 186         struct scsi_device *sdev;
 187         int total_failures = 0;
 188         int cmd_failed = 0;
 189         int cmd_cancel = 0;
 190         int devices_failed = 0;
 191
 192         shost_for_each_device(sdev, shost) {
 193                 list_for_each_entry(scmd, work_q, eh_entry) {
 194                         if (scmd->device == sdev) {
 195                                 ++total_failures;
 196                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
 197                                         ++cmd_cancel;
 198                                 else
 199                                         ++cmd_failed;
 200                         }
 201                 }
 202
 203                 if (cmd_cancel || cmd_failed) {
 204                         SCSI_LOG_ERROR_RECOVERY(3,
 205                                 sdev_printk(KERN_INFO, sdev,
 206                                             "%s: cmds failed: %d, cancel: %d\n",
 207                                             __func__, cmd_failed,
 208                                             cmd_cancel));
 209                         cmd_cancel = 0;
 210                         cmd_failed = 0;
 211                         ++devices_failed;
 212                 }
 213         }
 214
 215         SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
 216                                           " devices require eh work\n",
 217                                   total_failures, devices_failed));
 218 }
 219 #endif
 220
 221 /**
 222  * scsi_check_sense - Examine scsi cmd sense
 223  * @scmd:       Cmd to have sense checked.
 224  *
 225  * Return value:
 226  *      SUCCESS or FAILED or NEEDS_RETRY
 227  *
 228  * Notes:
 229  *      When a deferred error is detected the current command has
 230  *      not been executed and needs retrying.
 231  */
 232 static int scsi_check_sense(struct scsi_cmnd *scmd)
 233 {
 234         struct scsi_device *sdev = scmd->device;
 235         struct scsi_sense_hdr sshdr;
 236
 237         if (! scsi_command_normalize_sense(scmd, &sshdr))
 238                 return FAILED;  /* no valid sense data */
 239
 240         if (scsi_sense_is_deferred(&sshdr))
 241                 return NEEDS_RETRY;
 242
 243         if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
 244                         sdev->scsi_dh_data->scsi_dh->check_sense) {
 245                 int rc;
 246
 247                 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
 248                 if (rc != SCSI_RETURN_NOT_HANDLED)
 249                         return rc;
 250                 /* handler does not care. Drop down to default handling */
 251         }
 252
 253         /*
 254          * Previous logic looked for FILEMARK, EOM or ILI which are
 255          * mainly associated with tapes and returned SUCCESS.
 256          */
 257         if (sshdr.response_code == 0x70) {
 258                 /* fixed format */
 259                 if (scmd->sense_buffer[2] & 0xe0)
 260                         return SUCCESS;
 261         } else {
 262                 /*
 263                  * descriptor format: look for "stream commands sense data
 264                  * descriptor" (see SSC-3). Assume single sense data
 265                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
 266                  */
 267                 if ((sshdr.additional_length > 3) &&
 268                     (scmd->sense_buffer[8] == 0x4) &&
 269                     (scmd->sense_buffer[11] & 0xe0))
 270                         return SUCCESS;
 271         }
 272
 273         switch (sshdr.sense_key) {
 274         case NO_SENSE:
 275                 return SUCCESS;
 276         case RECOVERED_ERROR:
 277                 return /* soft_error */ SUCCESS;
 278
 279         case ABORTED_COMMAND:
 280                 if (sshdr.asc == 0x10) /* DIF */
 281                         return SUCCESS;
 282
 283                 return NEEDS_RETRY;
 284         case NOT_READY:
 285         case UNIT_ATTENTION:
 286                 /*
 287                  * if we are expecting a cc/ua because of a bus reset that we
 288                  * performed, treat this just as a retry.  otherwise this is
 289                  * information that we should pass up to the upper-level driver
 290                  * so that we can deal with it there.
 291                  */
 292                 if (scmd->device->expecting_cc_ua) {
 293                         scmd->device->expecting_cc_ua = 0;
 294                         return NEEDS_RETRY;
 295                 }
 296                 /*
 297                  * if the device is in the process of becoming ready, we
 298                  * should retry.
 299                  */
 300                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
 301                         return NEEDS_RETRY;
 302                 /*
 303                  * if the device is not started, we need to wake
 304                  * the error handler to start the motor
 305                  */
 306                 if (scmd->device->allow_restart &&
 307                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
 308                         return FAILED;
 309
 310                 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
 311                         scmd_printk(KERN_WARNING, scmd,
 312                                     "Warning! Received an indication that the "
 313                                     "LUN assignments on this target have "
 314                                     "changed. The Linux SCSI layer does not "
 315                                     "automatically remap LUN assignments.\n");
 316                 else if (sshdr.asc == 0x3f)
 317                         scmd_printk(KERN_WARNING, scmd,
 318                                     "Warning! Received an indication that the "
 319                                     "operating parameters on this target have "
 320                                     "changed. The Linux SCSI layer does not "
 321                                     "automatically adjust these parameters.\n");
 322
 323                 /*
 324                  * Pass the UA upwards for a determination in the completion
 325                  * functions.
 326                  */
 327                 return SUCCESS;
 328
 329                 /* these three are not supported */
 330         case COPY_ABORTED:
 331         case VOLUME_OVERFLOW:
 332         case MISCOMPARE:
 333                 return SUCCESS;
 334
 335         case MEDIUM_ERROR:
 336                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
 337                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
 338                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
 339                         return SUCCESS;
 340                 }
 341                 return NEEDS_RETRY;
 342
 343         case HARDWARE_ERROR:
 344                 if (scmd->device->retry_hwerror)
 345                         return ADD_TO_MLQUEUE;
 346                 else
 347                         return SUCCESS;
 348
 349         case ILLEGAL_REQUEST:
 350         case BLANK_CHECK:
 351         case DATA_PROTECT:
 352         default:
 353                 return SUCCESS;
 354         }
 355 }
 356
 357 static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
 358 {
 359         struct scsi_host_template *sht = sdev->host->hostt;
 360         struct scsi_device *tmp_sdev;
 361
 362         if (!sht->change_queue_depth ||
 363             sdev->queue_depth >= sdev->max_queue_depth)
 364                 return;
 365
 366         if (time_before(jiffies,
 367             sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
 368                 return;
 369
 370         if (time_before(jiffies,
 371             sdev->last_queue_full_time + sdev->queue_ramp_up_period))
 372                 return;
 373
 374         /*
 375          * Walk all devices of a target and do
 376          * ramp up on them.
 377          */
 378         shost_for_each_device(tmp_sdev, sdev->host) {
 379                 if (tmp_sdev->channel != sdev->channel ||
 380                     tmp_sdev->id != sdev->id ||
 381                     tmp_sdev->queue_depth == sdev->max_queue_depth)
 382                         continue;
 383                 /*
 384                  * call back into LLD to increase queue_depth by one
 385                  * with ramp up reason code.
 386                  */
 387                 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
 388                                         SCSI_QDEPTH_RAMP_UP);
 389                 sdev->last_queue_ramp_up = jiffies;
 390         }
 391 }
 392
 393 static void scsi_handle_queue_full(struct scsi_device *sdev)
 394 {
 395         struct scsi_host_template *sht = sdev->host->hostt;
 396         struct scsi_device *tmp_sdev;
 397
 398         if (!sht->change_queue_depth)
 399                 return;
 400
 401         shost_for_each_device(tmp_sdev, sdev->host) {
 402                 if (tmp_sdev->channel != sdev->channel ||
 403                     tmp_sdev->id != sdev->id)
 404                         continue;
 405                 /*
 406                  * We do not know the number of commands that were at
 407                  * the device when we got the queue full so we start
 408                  * from the highest possible value and work our way down.
 409                  */
 410                 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
 411                                         SCSI_QDEPTH_QFULL);
 412         }
 413 }
 414
 415 /**
 416  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
 417  * @scmd:       SCSI cmd to examine.
 418  *
 419  * Notes:
 420  *    This is *only* called when we are examining the status of commands
 421  *    queued during error recovery.  the main difference here is that we
 422  *    don't allow for the possibility of retries here, and we are a lot
 423  *    more restrictive about what we consider acceptable.
 424  */
 425 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
 426 {
 427         /*
 428          * first check the host byte, to see if there is anything in there
 429          * that would indicate what we need to do.
 430          */
 431         if (host_byte(scmd->result) == DID_RESET) {
 432                 /*
 433                  * rats.  we are already in the error handler, so we now
 434                  * get to try and figure out what to do next.  if the sense
 435                  * is valid, we have a pretty good idea of what to do.
 436                  * if not, we mark it as FAILED.
 437                  */
 438                 return scsi_check_sense(scmd);
 439         }
 440         if (host_byte(scmd->result) != DID_OK)
 441                 return FAILED;
 442
 443         /*
 444          * next, check the message byte.
 445          */
 446         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
 447                 return FAILED;
 448
 449         /*
 450          * now, check the status byte to see if this indicates
 451          * anything special.
 452          */
 453         switch (status_byte(scmd->result)) {
 454         case GOOD:
 455                 scsi_handle_queue_ramp_up(scmd->device);
 456         case COMMAND_TERMINATED:
 457                 return SUCCESS;
 458         case CHECK_CONDITION:
 459                 return scsi_check_sense(scmd);
 460         case CONDITION_GOOD:
 461         case INTERMEDIATE_GOOD:
 462         case INTERMEDIATE_C_GOOD:
 463                 /*
 464                  * who knows?  FIXME(eric)
 465                  */
 466                 return SUCCESS;
 467         case RESERVATION_CONFLICT:
 468                 if (scmd->cmnd[0] == TEST_UNIT_READY)
 469                         /* it is a success, we probed the device and
 470                          * found it */
 471                         return SUCCESS;
 472                 /* otherwise, we failed to send the command */
 473                 return FAILED;
 474         case QUEUE_FULL:
 475                 scsi_handle_queue_full(scmd->device);
 476                 /* fall through */
 477         case BUSY:
 478                 return NEEDS_RETRY;
 479         default:
 480                 return FAILED;
 481         }
 482         return FAILED;
 483 }
 484
 485 /**
 486  * scsi_eh_done - Completion function for error handling.
 487  * @scmd:       Cmd that is done.
 488  */
 489 static void scsi_eh_done(struct scsi_cmnd *scmd)
 490 {
 491         struct completion     *eh_action;
 492
 493         SCSI_LOG_ERROR_RECOVERY(3,
 494                 printk("%s scmd: %p result: %x\n",
 495                         __func__, scmd, scmd->result));
 496
 497         eh_action = scmd->device->host->eh_action;
 498         if (eh_action)
 499                 complete(eh_action);
 500 }
 501
 502 /**
 503  * scsi_try_host_reset - ask host adapter to reset itself
 504  * @scmd:       SCSI cmd to send hsot reset.
 505  */
 506 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
 507 {
 508         unsigned long flags;
 509         int rtn;
 510
 511         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
 512                                           __func__));
 513
 514         if (!scmd->device->host->hostt->eh_host_reset_handler)
 515                 return FAILED;
 516
 517         rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
 518
 519         if (rtn == SUCCESS) {
 520                 if (!scmd->device->host->hostt->skip_settle_delay)
 521                         ssleep(HOST_RESET_SETTLE_TIME);
 522                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
 523                 scsi_report_bus_reset(scmd->device->host,
 524                                       scmd_channel(scmd));
 525                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
 526         }
 527
 528         return rtn;
 529 }
 530
 531 /**
 532  * scsi_try_bus_reset - ask host to perform a bus reset
 533  * @scmd:       SCSI cmd to send bus reset.
 534  */
 535 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
 536 {
 537         unsigned long flags;
 538         int rtn;
 539
 540         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
 541                                           __func__));
 542
 543         if (!scmd->device->host->hostt->eh_bus_reset_handler)
 544                 return FAILED;
 545
 546         rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
 547
 548         if (rtn == SUCCESS) {
 549                 if (!scmd->device->host->hostt->skip_settle_delay)
 550                         ssleep(BUS_RESET_SETTLE_TIME);
 551                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
 552                 scsi_report_bus_reset(scmd->device->host,
 553                                       scmd_channel(scmd));
 554                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
 555         }
 556
 557         return rtn;
 558 }
 559
 560 static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
 561 {
 562         sdev->was_reset = 1;
 563         sdev->expecting_cc_ua = 1;
 564 }
 565
 566 /**
 567  * scsi_try_target_reset - Ask host to perform a target reset
 568  * @scmd:       SCSI cmd used to send a target reset
 569  *
 570  * Notes:
 571  *    There is no timeout for this operation.  if this operation is
 572  *    unreliable for a given host, then the host itself needs to put a
 573  *    timer on it, and set the host back to a consistent state prior to
 574  *    returning.
 575  */
 576 static int scsi_try_target_reset(struct scsi_cmnd *scmd)
 577 {
 578         unsigned long flags;
 579         int rtn;
 580
 581         if (!scmd->device->host->hostt->eh_target_reset_handler)
 582                 return FAILED;
 583
 584         rtn = scmd->device->host->hostt->eh_target_reset_handler(scmd);
 585         if (rtn == SUCCESS) {
 586                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
 587                 __starget_for_each_device(scsi_target(scmd->device), NULL,
 588                                           __scsi_report_device_reset);
 589                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
 590         }
 591
 592         return rtn;
 593 }
 594
 595 /**
 596  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
 597  * @scmd:       SCSI cmd used to send BDR
 598  *
 599  * Notes:
 600  *    There is no timeout for this operation.  if this operation is
 601  *    unreliable for a given host, then the host itself needs to put a
 602  *    timer on it, and set the host back to a consistent state prior to
 603  *    returning.
 604  */
 605 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
 606 {
 607         int rtn;
 608
 609         if (!scmd->device->host->hostt->eh_device_reset_handler)
 610                 return FAILED;
 611
 612         rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
 613         if (rtn == SUCCESS)
 614                 __scsi_report_device_reset(scmd->device, NULL);
 615         return rtn;
 616 }
 617
 618 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
 619 {
 620         if (!scmd->device->host->hostt->eh_abort_handler)
 621                 return FAILED;
 622
 623         return scmd->device->host->hostt->eh_abort_handler(scmd);
 624 }
 625
 626 /**
 627  * scsi_try_to_abort_cmd - Ask host to abort a running command.
 628  * @scmd:       SCSI cmd to abort from Lower Level.
 629  *
 630  * Notes:
 631  *    This function will not return until the user's completion function
 632  *    has been called.  there is no timeout on this operation.  if the
 633  *    author of the low-level driver wishes this operation to be timed,
 634  *    they can provide this facility themselves.  helper functions in
 635  *    scsi_error.c can be supplied to make this easier to do.
 636  */
 637 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
 638 {
 639         /*
 640          * scsi_done was called just after the command timed out and before
 641          * we had a chance to process it. (db)
 642          */
 643         if (scmd->serial_number == 0)
 644                 return SUCCESS;
 645         return __scsi_try_to_abort_cmd(scmd);
 646 }
 647
 648 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
 649 {
 650         if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
 651                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
 652                         if (scsi_try_target_reset(scmd) != SUCCESS)
 653                                 if (scsi_try_bus_reset(scmd) != SUCCESS)
 654                                         scsi_try_host_reset(scmd);
 655 }
 656
 657 /**
 658  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recory
 659  * @scmd:       SCSI command structure to hijack
 660  * @ses:        structure to save restore information
 661  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
 662  * @cmnd_size:  size in bytes of @cmnd (must be <= BLK_MAX_CDB)
 663  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
 664  *
 665  * This function is used to save a scsi command information before re-execution
 666  * as part of the error recovery process.  If @sense_bytes is 0 the command
 667  * sent must be one that does not transfer any data.  If @sense_bytes != 0
 668  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
 669  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
 670  */
 671 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
 672                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
 673 {
 674         struct scsi_device *sdev = scmd->device;
 675
 676         /*
 677          * We need saved copies of a number of fields - this is because
 678          * error handling may need to overwrite these with different values
 679          * to run different commands, and once error handling is complete,
 680          * we will need to restore these values prior to running the actual
 681          * command.
 682          */
 683         ses->cmd_len = scmd->cmd_len;
 684         ses->cmnd = scmd->cmnd;
 685         ses->data_direction = scmd->sc_data_direction;
 686         ses->sdb = scmd->sdb;
 687         ses->next_rq = scmd->request->next_rq;
 688         ses->result = scmd->result;
 689         ses->underflow = scmd->underflow;
 690         ses->prot_op = scmd->prot_op;
 691
 692         scmd->prot_op = SCSI_PROT_NORMAL;
 693         scmd->cmnd = ses->eh_cmnd;
 694         memset(scmd->cmnd, 0, BLK_MAX_CDB);
 695         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
 696         scmd->request->next_rq = NULL;
 697
 698         if (sense_bytes) {
 699                 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
 700                                          sense_bytes);
 701                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
 702                             scmd->sdb.length);
 703                 scmd->sdb.table.sgl = &ses->sense_sgl;
 704                 scmd->sc_data_direction = DMA_FROM_DEVICE;
 705                 scmd->sdb.table.nents = 1;
 706                 scmd->cmnd[0] = REQUEST_SENSE;
 707                 scmd->cmnd[4] = scmd->sdb.length;
 708                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 709         } else {
 710                 scmd->sc_data_direction = DMA_NONE;
 711                 if (cmnd) {
 712                         BUG_ON(cmnd_size > BLK_MAX_CDB);
 713                         memcpy(scmd->cmnd, cmnd, cmnd_size);
 714                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
 715                 }
 716         }
 717
 718         scmd->underflow = 0;
 719
 720         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
 721                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
 722                         (sdev->lun << 5 & 0xe0);
 723
 724         /*
 725          * Zero the sense buffer.  The scsi spec mandates that any
 726          * untransferred sense data should be interpreted as being zero.
 727          */
 728         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 729 }
 730 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
 731
 732 /**
 733  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recory
 734  * @scmd:       SCSI command structure to restore
 735  * @ses:        saved information from a coresponding call to scsi_eh_prep_cmnd
 736  *
 737  * Undo any damage done by above scsi_eh_prep_cmnd().
 738  */
 739 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
 740 {
 741         /*
 742          * Restore original data
 743          */
 744         scmd->cmd_len = ses->cmd_len;
 745         scmd->cmnd = ses->cmnd;
 746         scmd->sc_data_direction = ses->data_direction;
 747         scmd->sdb = ses->sdb;
 748         scmd->request->next_rq = ses->next_rq;
 749         scmd->result = ses->result;
 750         scmd->underflow = ses->underflow;
 751         scmd->prot_op = ses->prot_op;
 752 }
 753 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
 754
 755 /**
 756  * scsi_send_eh_cmnd  - submit a scsi command as part of error recory
 757  * @scmd:       SCSI command structure to hijack
 758  * @cmnd:       CDB to send
 759  * @cmnd_size:  size in bytes of @cmnd
 760  * @timeout:    timeout for this request
 761  * @sense_bytes: size of sense data to copy or 0
 762  *
 763  * This function is used to send a scsi command down to a target device
 764  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
 765  *
 766  * Return value:
 767  *    SUCCESS or FAILED or NEEDS_RETRY
 768  */
 769 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
 770                              int cmnd_size, int timeout, unsigned sense_bytes)
 771 {
 772         struct scsi_device *sdev = scmd->device;
 773         struct Scsi_Host *shost = sdev->host;
 774         DECLARE_COMPLETION_ONSTACK(done);
 775         unsigned long timeleft;
 776         unsigned long flags;
 777         struct scsi_eh_save ses;
 778         int rtn;
 779
 780         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
 781         shost->eh_action = &done;
 782
 783         spin_lock_irqsave(shost->host_lock, flags);
 784         scsi_log_send(scmd);
 785         shost->hostt->queuecommand(scmd, scsi_eh_done);
 786         spin_unlock_irqrestore(shost->host_lock, flags);
 787
 788         timeleft = wait_for_completion_timeout(&done, timeout);
 789
 790         shost->eh_action = NULL;
 791
 792         scsi_log_completion(scmd, SUCCESS);
 793
 794         SCSI_LOG_ERROR_RECOVERY(3,
 795                 printk("%s: scmd: %p, timeleft: %ld\n",
 796                         __func__, scmd, timeleft));
 797
 798         /*
 799          * If there is time left scsi_eh_done got called, and we will
 800          * examine the actual status codes to see whether the command
 801          * actually did complete normally, else tell the host to forget
 802          * about this command.
 803          */
 804         if (timeleft) {
 805                 rtn = scsi_eh_completed_normally(scmd);
 806                 SCSI_LOG_ERROR_RECOVERY(3,
 807                         printk("%s: scsi_eh_completed_normally %x\n",
 808                                __func__, rtn));
 809
 810                 switch (rtn) {
 811                 case SUCCESS:
 812                 case NEEDS_RETRY:
 813                 case FAILED:
 814                         break;
 815                 case ADD_TO_MLQUEUE:
 816                         rtn = NEEDS_RETRY;
 817                         break;
 818                 default:
 819                         rtn = FAILED;
 820                         break;
 821                 }
 822         } else {
 823                 scsi_abort_eh_cmnd(scmd);
 824                 rtn = FAILED;
 825         }
 826
 827         scsi_eh_restore_cmnd(scmd, &ses);
 828         return rtn;
 829 }
 830
 831 /**
 832  * scsi_request_sense - Request sense data from a particular target.
 833  * @scmd:       SCSI cmd for request sense.
 834  *
 835  * Notes:
 836  *    Some hosts automatically obtain this information, others require
 837  *    that we obtain it on our own. This function will *not* return until
 838  *    the command either times out, or it completes.
 839  */
 840 static int scsi_request_sense(struct scsi_cmnd *scmd)
 841 {
 842         return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
 843 }
 844
 845 /**
 846  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
 847  * @scmd:       Original SCSI cmd that eh has finished.
 848  * @done_q:     Queue for processed commands.
 849  *
 850  * Notes:
 851  *    We don't want to use the normal command completion while we are are
 852  *    still handling errors - it may cause other commands to be queued,
 853  *    and that would disturb what we are doing.  Thus we really want to
 854  *    keep a list of pending commands for final completion, and once we
 855  *    are ready to leave error handling we handle completion for real.
 856  */
 857 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
 858 {
 859         scmd->device->host->host_failed--;
 860         scmd->eh_eflags = 0;
 861         list_move_tail(&scmd->eh_entry, done_q);
 862 }
 863 EXPORT_SYMBOL(scsi_eh_finish_cmd);
 864
 865 /**
 866  * scsi_eh_get_sense - Get device sense data.
 867  * @work_q:     Queue of commands to process.
 868  * @done_q:     Queue of processed commands.
 869  *
 870  * Description:
 871  *    See if we need to request sense information.  if so, then get it
 872  *    now, so we have a better idea of what to do.
 873  *
 874  * Notes:
 875  *    This has the unfortunate side effect that if a shost adapter does
 876  *    not automatically request sense information, we end up shutting
 877  *    it down before we request it.
 878  *
 879  *    All drivers should request sense information internally these days,
 880  *    so for now all I have to say is tough noogies if you end up in here.
 881  *
 882  *    XXX: Long term this code should go away, but that needs an audit of
 883  *         all LLDDs first.
 884  */
 885 int scsi_eh_get_sense(struct list_head *work_q,
 886                       struct list_head *done_q)
 887 {
 888         struct scsi_cmnd *scmd, *next;
 889         int rtn;
 890
 891         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
 892                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
 893                     SCSI_SENSE_VALID(scmd))
 894                         continue;
 895
 896                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
 897                                                   "%s: requesting sense\n",
 898                                                   current->comm));
 899                 rtn = scsi_request_sense(scmd);
 900                 if (rtn != SUCCESS)
 901                         continue;
 902
 903                 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
 904                                                   " result %x\n", scmd,
 905                                                   scmd->result));
 906                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
 907
 908                 rtn = scsi_decide_disposition(scmd);
 909
 910                 /*
 911                  * if the result was normal, then just pass it along to the
 912                  * upper level.
 913                  */
 914                 if (rtn == SUCCESS)
 915                         /* we don't want this command reissued, just
 916                          * finished with the sense data, so set
 917                          * retries to the max allowed to ensure it
 918                          * won't get reissued */
 919                         scmd->retries = scmd->allowed;
 920                 else if (rtn != NEEDS_RETRY)
 921                         continue;
 922
 923                 scsi_eh_finish_cmd(scmd, done_q);
 924         }
 925
 926         return list_empty(work_q);
 927 }
 928 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
 929
 930 /**
 931  * scsi_eh_tur - Send TUR to device.
 932  * @scmd:       &scsi_cmnd to send TUR
 933  *
 934  * Return value:
 935  *    0 - Device is ready. 1 - Device NOT ready.
 936  */
 937 static int scsi_eh_tur(struct scsi_cmnd *scmd)
 938 {
 939         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
 940         int retry_cnt = 1, rtn;
 941
 942 retry_tur:
 943         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
 944
 945         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
 946                 __func__, scmd, rtn));
 947
 948         switch (rtn) {
 949         case NEEDS_RETRY:
 950                 if (retry_cnt--)
 951                         goto retry_tur;
 952                 /*FALLTHRU*/
 953         case SUCCESS:
 954                 return 0;
 955         default:
 956                 return 1;
 957         }
 958 }
 959
 960 /**
 961  * scsi_eh_abort_cmds - abort pending commands.
 962  * @work_q:     &list_head for pending commands.
 963  * @done_q:     &list_head for processed commands.
 964  *
 965  * Decription:
 966  *    Try and see whether or not it makes sense to try and abort the
 967  *    running command.  This only works out to be the case if we have one
 968  *    command that has timed out.  If the command simply failed, it makes
 969  *    no sense to try and abort the command, since as far as the shost
 970  *    adapter is concerned, it isn't running.
 971  */
 972 static int scsi_eh_abort_cmds(struct list_head *work_q,
 973                               struct list_head *done_q)
 974 {
 975         struct scsi_cmnd *scmd, *next;
 976         int rtn;
 977
 978         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
 979                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
 980                         continue;
 981                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
 982                                                   "0x%p\n", current->comm,
 983                                                   scmd));
 984                 rtn = scsi_try_to_abort_cmd(scmd);
 985                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
 986                         scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
 987                         if (!scsi_device_online(scmd->device) ||
 988                             rtn == FAST_IO_FAIL ||
 989                             !scsi_eh_tur(scmd)) {
 990                                 scsi_eh_finish_cmd(scmd, done_q);
 991                         }
 992
 993                 } else
 994                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
 995                                                           " cmd failed:"
 996                                                           "0x%p\n",
 997                                                           current->comm,
 998                                                           scmd));
 999         }
1000
1001         return list_empty(work_q);
1002 }
1003
1004 /**
1005  * scsi_eh_try_stu - Send START_UNIT to device.
1006  * @scmd:       &scsi_cmnd to send START_UNIT
1007  *
1008  * Return value:
1009  *    0 - Device is ready. 1 - Device NOT ready.
1010  */
1011 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1012 {
1013         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
1014
1015         if (scmd->device->allow_restart) {
1016                 int i, rtn = NEEDS_RETRY;
1017
1018                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
1019                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
1020
1021                 if (rtn == SUCCESS)
1022                         return 0;
1023         }
1024
1025         return 1;
1026 }
1027
1028  /**
1029  * scsi_eh_stu - send START_UNIT if needed
1030  * @shost:      &scsi host being recovered.
1031  * @work_q:     &list_head for pending commands.
1032  * @done_q:     &list_head for processed commands.
1033  *
1034  * Notes:
1035  *    If commands are failing due to not ready, initializing command required,
1036  *      try revalidating the device, which will end up sending a start unit.
1037  */
1038 static int scsi_eh_stu(struct Scsi_Host *shost,
1039                               struct list_head *work_q,
1040                               struct list_head *done_q)
1041 {
1042         struct scsi_cmnd *scmd, *stu_scmd, *next;
1043         struct scsi_device *sdev;
1044
1045         shost_for_each_device(sdev, shost) {
1046                 stu_scmd = NULL;
1047                 list_for_each_entry(scmd, work_q, eh_entry)
1048                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1049                             scsi_check_sense(scmd) == FAILED ) {
1050                                 stu_scmd = scmd;
1051                                 break;
1052                         }
1053
1054                 if (!stu_scmd)
1055                         continue;
1056
1057                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1058                                                   " 0x%p\n", current->comm, sdev));
1059
1060                 if (!scsi_eh_try_stu(stu_scmd)) {
1061                         if (!scsi_device_online(sdev) ||
1062                             !scsi_eh_tur(stu_scmd)) {
1063                                 list_for_each_entry_safe(scmd, next,
1064                                                           work_q, eh_entry) {
1065                                         if (scmd->device == sdev)
1066                                                 scsi_eh_finish_cmd(scmd, done_q);
1067                                 }
1068                         }
1069                 } else {
1070                         SCSI_LOG_ERROR_RECOVERY(3,
1071                                                 printk("%s: START_UNIT failed to sdev:"
1072                                                        " 0x%p\n", current->comm, sdev));
1073                 }
1074         }
1075
1076         return list_empty(work_q);
1077 }
1078
1079
1080 /**
1081  * scsi_eh_bus_device_reset - send bdr if needed
1082  * @shost:      scsi host being recovered.
1083  * @work_q:     &list_head for pending commands.
1084  * @done_q:     &list_head for processed commands.
1085  *
1086  * Notes:
1087  *    Try a bus device reset.  Still, look to see whether we have multiple
1088  *    devices that are jammed or not - if we have multiple devices, it
1089  *    makes no sense to try bus_device_reset - we really would need to try
1090  *    a bus_reset instead.
1091  */
1092 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1093                                     struct list_head *work_q,
1094                                     struct list_head *done_q)
1095 {
1096         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1097         struct scsi_device *sdev;
1098         int rtn;
1099
1100         shost_for_each_device(sdev, shost) {
1101                 bdr_scmd = NULL;
1102                 list_for_each_entry(scmd, work_q, eh_entry)
1103                         if (scmd->device == sdev) {
1104                                 bdr_scmd = scmd;
1105                                 break;
1106                         }
1107
1108                 if (!bdr_scmd)
1109                         continue;
1110
1111                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1112                                                   " 0x%p\n", current->comm,
1113                                                   sdev));
1114                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1115                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1116                         if (!scsi_device_online(sdev) ||
1117                             rtn == FAST_IO_FAIL ||
1118                             !scsi_eh_tur(bdr_scmd)) {
1119                                 list_for_each_entry_safe(scmd, next,
1120                                                          work_q, eh_entry) {
1121                                         if (scmd->device == sdev)
1122                                                 scsi_eh_finish_cmd(scmd,
1123                                                                    done_q);
1124                                 }
1125                         }
1126                 } else {
1127                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1128                                                           " failed sdev:"
1129                                                           "0x%p\n",
1130                                                           current->comm,
1131                                                            sdev));
1132                 }
1133         }
1134
1135         return list_empty(work_q);
1136 }
1137
1138 /**
1139  * scsi_eh_target_reset - send target reset if needed
1140  * @shost:      scsi host being recovered.
1141  * @work_q:     &list_head for pending commands.
1142  * @done_q:     &list_head for processed commands.
1143  *
1144  * Notes:
1145  *    Try a target reset.
1146  */
1147 static int scsi_eh_target_reset(struct Scsi_Host *shost,
1148                                 struct list_head *work_q,
1149                                 struct list_head *done_q)
1150 {
1151         struct scsi_cmnd *scmd, *tgtr_scmd, *next;
1152         unsigned int id = 0;
1153         int rtn;
1154
1155         do {
1156                 tgtr_scmd = NULL;
1157                 list_for_each_entry(scmd, work_q, eh_entry) {
1158                         if (id == scmd_id(scmd)) {
1159                                 tgtr_scmd = scmd;
1160                                 break;
1161                         }
1162                 }
1163                 if (!tgtr_scmd) {
1164                         /* not one exactly equal; find the next highest */
1165                         list_for_each_entry(scmd, work_q, eh_entry) {
1166                                 if (scmd_id(scmd) > id &&
1167                                     (!tgtr_scmd ||
1168                                      scmd_id(tgtr_scmd) > scmd_id(scmd)))
1169                                                 tgtr_scmd = scmd;
1170                         }
1171                 }
1172                 if (!tgtr_scmd)
1173                         /* no more commands, that's it */
1174                         break;
1175
1176                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1177                                                   "to target %d\n",
1178                                                   current->comm, id));
1179                 rtn = scsi_try_target_reset(tgtr_scmd);
1180                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1181                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1182                                 if (id == scmd_id(scmd))
1183                                         if (!scsi_device_online(scmd->device) ||
1184                                             rtn == FAST_IO_FAIL ||
1185                                             !scsi_eh_tur(tgtr_scmd))
1186                                                 scsi_eh_finish_cmd(scmd,
1187                                                                    done_q);
1188                         }
1189                 } else
1190                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1191                                                           " failed target: "
1192                                                           "%d\n",
1193                                                           current->comm, id));
1194                 id++;
1195         } while(id != 0);
1196
1197         return list_empty(work_q);
1198 }
1199
1200 /**
1201  * scsi_eh_bus_reset - send a bus reset
1202  * @shost:      &scsi host being recovered.
1203  * @work_q:     &list_head for pending commands.
1204  * @done_q:     &list_head for processed commands.
1205  */
1206 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1207                              struct list_head *work_q,
1208                              struct list_head *done_q)
1209 {
1210         struct scsi_cmnd *scmd, *chan_scmd, *next;
1211         unsigned int channel;
1212         int rtn;
1213
1214         /*
1215          * we really want to loop over the various channels, and do this on
1216          * a channel by channel basis.  we should also check to see if any
1217          * of the failed commands are on soft_reset devices, and if so, skip
1218          * the reset.
1219          */
1220
1221         for (channel = 0; channel <= shost->max_channel; channel++) {
1222                 chan_scmd = NULL;
1223                 list_for_each_entry(scmd, work_q, eh_entry) {
1224                         if (channel == scmd_channel(scmd)) {
1225                                 chan_scmd = scmd;
1226                                 break;
1227                                 /*
1228                                  * FIXME add back in some support for
1229                                  * soft_reset devices.
1230                                  */
1231                         }
1232                 }
1233
1234                 if (!chan_scmd)
1235                         continue;
1236                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1237                                                   " %d\n", current->comm,
1238                                                   channel));
1239                 rtn = scsi_try_bus_reset(chan_scmd);
1240                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1241                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1242                                 if (channel == scmd_channel(scmd))
1243                                         if (!scsi_device_online(scmd->device) ||
1244                                             rtn == FAST_IO_FAIL ||
1245                                             !scsi_eh_tur(scmd))
1246                                                 scsi_eh_finish_cmd(scmd,
1247                                                                    done_q);
1248                         }
1249                 } else {
1250                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1251                                                           " failed chan: %d\n",
1252                                                           current->comm,
1253                                                           channel));
1254                 }
1255         }
1256         return list_empty(work_q);
1257 }
1258
1259 /**
1260  * scsi_eh_host_reset - send a host reset
1261  * @work_q:     list_head for processed commands.
1262  * @done_q:     list_head for processed commands.
1263  */
1264 static int scsi_eh_host_reset(struct list_head *work_q,
1265                               struct list_head *done_q)
1266 {
1267         struct scsi_cmnd *scmd, *next;
1268         int rtn;
1269
1270         if (!list_empty(work_q)) {
1271                 scmd = list_entry(work_q->next,
1272                                   struct scsi_cmnd, eh_entry);
1273
1274                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1275                                                   , current->comm));
1276
1277                 rtn = scsi_try_host_reset(scmd);
1278                 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
1279                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1280                                 if (!scsi_device_online(scmd->device) ||
1281                                     rtn == FAST_IO_FAIL ||
1282                                     (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1283                                     !scsi_eh_tur(scmd))
1284                                         scsi_eh_finish_cmd(scmd, done_q);
1285                         }
1286                 } else {
1287                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1288                                                           " failed\n",
1289                                                           current->comm));
1290                 }
1291         }
1292         return list_empty(work_q);
1293 }
1294
1295 /**
1296  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1297  * @work_q:     list_head for processed commands.
1298  * @done_q:     list_head for processed commands.
1299  */
1300 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1301                                   struct list_head *done_q)
1302 {
1303         struct scsi_cmnd *scmd, *next;
1304
1305         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1306                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1307                             "not ready after error recovery\n");
1308                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1309                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1310                         /*
1311                          * FIXME: Handle lost cmds.
1312                          */
1313                 }
1314                 scsi_eh_finish_cmd(scmd, done_q);
1315         }
1316         return;
1317 }
1318
1319 /**
1320  * scsi_noretry_cmd - determinte if command should be failed fast
1321  * @scmd:       SCSI cmd to examine.
1322  */
1323 int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1324 {
1325         switch (host_byte(scmd->result)) {
1326         case DID_OK:
1327                 break;
1328         case DID_BUS_BUSY:
1329                 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
1330         case DID_PARITY:
1331                 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
1332         case DID_ERROR:
1333                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1334                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1335                         return 0;
1336                 /* fall through */
1337         case DID_SOFT_ERROR:
1338                 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
1339         }
1340
1341         switch (status_byte(scmd->result)) {
1342         case CHECK_CONDITION:
1343                 /*
1344                  * assume caller has checked sense and determinted
1345                  * the check condition was retryable.
1346                  */
1347                 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1348                     scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1349                         return 1;
1350         }
1351
1352         return 0;
1353 }
1354
1355 /**
1356  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1357  * @scmd:       SCSI cmd to examine.
1358  *
1359  * Notes:
1360  *    This is *only* called when we are examining the status after sending
1361  *    out the actual data command.  any commands that are queued for error
1362  *    recovery (e.g. test_unit_ready) do *not* come through here.
1363  *
1364  *    When this routine returns failed, it means the error handler thread
1365  *    is woken.  In cases where the error code indicates an error that
1366  *    doesn't require the error handler read (i.e. we don't need to
1367  *    abort/reset), this function should return SUCCESS.
1368  */
1369 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1370 {
1371         int rtn;
1372
1373         /*
1374          * if the device is offline, then we clearly just pass the result back
1375          * up to the top level.
1376          */
1377         if (!scsi_device_online(scmd->device)) {
1378                 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1379                                                   " as SUCCESS\n",
1380                                                   __func__));
1381                 return SUCCESS;
1382         }
1383
1384         /*
1385          * first check the host byte, to see if there is anything in there
1386          * that would indicate what we need to do.
1387          */
1388         switch (host_byte(scmd->result)) {
1389         case DID_PASSTHROUGH:
1390                 /*
1391                  * no matter what, pass this through to the upper layer.
1392                  * nuke this special code so that it looks like we are saying
1393                  * did_ok.
1394                  */
1395                 scmd->result &= 0xff00ffff;
1396                 return SUCCESS;
1397         case DID_OK:
1398                 /*
1399                  * looks good.  drop through, and check the next byte.
1400                  */
1401                 break;
1402         case DID_NO_CONNECT:
1403         case DID_BAD_TARGET:
1404         case DID_ABORT:
1405                 /*
1406                  * note - this means that we just report the status back
1407                  * to the top level driver, not that we actually think
1408                  * that it indicates SUCCESS.
1409                  */
1410                 return SUCCESS;
1411                 /*
1412                  * when the low level driver returns did_soft_error,
1413                  * it is responsible for keeping an internal retry counter
1414                  * in order to avoid endless loops (db)
1415                  *
1416                  * actually this is a bug in this function here.  we should
1417                  * be mindful of the maximum number of retries specified
1418                  * and not get stuck in a loop.
1419                  */
1420         case DID_SOFT_ERROR:
1421                 goto maybe_retry;
1422         case DID_IMM_RETRY:
1423                 return NEEDS_RETRY;
1424
1425         case DID_REQUEUE:
1426                 return ADD_TO_MLQUEUE;
1427         case DID_TRANSPORT_DISRUPTED:
1428                 /*
1429                  * LLD/transport was disrupted during processing of the IO.
1430                  * The transport class is now blocked/blocking,
1431                  * and the transport will decide what to do with the IO
1432                  * based on its timers and recovery capablilities if
1433                  * there are enough retries.
1434                  */
1435                 goto maybe_retry;
1436         case DID_TRANSPORT_FAILFAST:
1437                 /*
1438                  * The transport decided to failfast the IO (most likely
1439                  * the fast io fail tmo fired), so send IO directly upwards.
1440                  */
1441                 return SUCCESS;
1442         case DID_ERROR:
1443                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1444                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1445                         /*
1446                          * execute reservation conflict processing code
1447                          * lower down
1448                          */
1449                         break;
1450                 /* fallthrough */
1451
1452         case DID_BUS_BUSY:
1453         case DID_PARITY:
1454                 goto maybe_retry;
1455         case DID_TIME_OUT:
1456                 /*
1457                  * when we scan the bus, we get timeout messages for
1458                  * these commands if there is no device available.
1459                  * other hosts report did_no_connect for the same thing.
1460                  */
1461                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1462                      scmd->cmnd[0] == INQUIRY)) {
1463                         return SUCCESS;
1464                 } else {
1465                         return FAILED;
1466                 }
1467         case DID_RESET:
1468                 return SUCCESS;
1469         default:
1470                 return FAILED;
1471         }
1472
1473         /*
1474          * next, check the message byte.
1475          */
1476         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1477                 return FAILED;
1478
1479         /*
1480          * check the status byte to see if this indicates anything special.
1481          */
1482         switch (status_byte(scmd->result)) {
1483         case QUEUE_FULL:
1484                 scsi_handle_queue_full(scmd->device);
1485                 /*
1486                  * the case of trying to send too many commands to a
1487                  * tagged queueing device.
1488                  */
1489         case BUSY:
1490                 /*
1491                  * device can't talk to us at the moment.  Should only
1492                  * occur (SAM-3) when the task queue is empty, so will cause
1493                  * the empty queue handling to trigger a stall in the
1494                  * device.
1495                  */
1496                 return ADD_TO_MLQUEUE;
1497         case GOOD:
1498                 scsi_handle_queue_ramp_up(scmd->device);
1499         case COMMAND_TERMINATED:
1500                 return SUCCESS;
1501         case TASK_ABORTED:
1502                 goto maybe_retry;
1503         case CHECK_CONDITION:
1504                 rtn = scsi_check_sense(scmd);
1505                 if (rtn == NEEDS_RETRY)
1506                         goto maybe_retry;
1507                 /* if rtn == FAILED, we have no sense information;
1508                  * returning FAILED will wake the error handler thread
1509                  * to collect the sense and redo the decide
1510                  * disposition */
1511                 return rtn;
1512         case CONDITION_GOOD:
1513         case INTERMEDIATE_GOOD:
1514         case INTERMEDIATE_C_GOOD:
1515         case ACA_ACTIVE:
1516                 /*
1517                  * who knows?  FIXME(eric)
1518                  */
1519                 return SUCCESS;
1520
1521         case RESERVATION_CONFLICT:
1522                 sdev_printk(KERN_INFO, scmd->device,
1523                             "reservation conflict\n");
1524                 return SUCCESS; /* causes immediate i/o error */
1525         default:
1526                 return FAILED;
1527         }
1528         return FAILED;
1529
1530       maybe_retry:
1531
1532         /* we requeue for retry because the error was retryable, and
1533          * the request was not marked fast fail.  Note that above,
1534          * even if the request is marked fast fail, we still requeue
1535          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1536         if ((++scmd->retries) <= scmd->allowed
1537             && !scsi_noretry_cmd(scmd)) {
1538                 return NEEDS_RETRY;
1539         } else {
1540                 /*
1541                  * no more retries - report this one back to upper level.
1542                  */
1543                 return SUCCESS;
1544         }
1545 }
1546
1547 static void eh_lock_door_done(struct request *req, int uptodate)
1548 {
1549         __blk_put_request(req->q, req);
1550 }
1551
1552 /**
1553  * scsi_eh_lock_door - Prevent medium removal for the specified device
1554  * @sdev:       SCSI device to prevent medium removal
1555  *
1556  * Locking:
1557  *      We must be called from process context.
1558  *
1559  * Notes:
1560  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1561  *      head of the devices request queue, and continue.
1562  */
1563 static void scsi_eh_lock_door(struct scsi_device *sdev)
1564 {
1565         struct request *req;
1566
1567         /*
1568          * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1569          * request becomes available
1570          */
1571         req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
1572
1573         req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1574         req->cmd[1] = 0;
1575         req->cmd[2] = 0;
1576         req->cmd[3] = 0;
1577         req->cmd[4] = SCSI_REMOVAL_PREVENT;
1578         req->cmd[5] = 0;
1579
1580         req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1581
1582         req->cmd_type = REQ_TYPE_BLOCK_PC;
1583         req->cmd_flags |= REQ_QUIET;
1584         req->timeout = 10 * HZ;
1585         req->retries = 5;
1586
1587         blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
1588 }
1589
1590 /**
1591  * scsi_restart_operations - restart io operations to the specified host.
1592  * @shost:      Host we are restarting.
1593  *
1594  * Notes:
1595  *    When we entered the error handler, we blocked all further i/o to
1596  *    this device.  we need to 'reverse' this process.
1597  */
1598 static void scsi_restart_operations(struct Scsi_Host *shost)
1599 {
1600         struct scsi_device *sdev;
1601         unsigned long flags;
1602
1603         /*
1604          * If the door was locked, we need to insert a door lock request
1605          * onto the head of the SCSI request queue for the device.  There
1606          * is no point trying to lock the door of an off-line device.
1607          */
1608         shost_for_each_device(sdev, shost) {
1609                 if (scsi_device_online(sdev) && sdev->locked)
1610                         scsi_eh_lock_door(sdev);
1611         }
1612
1613         /*
1614          * next free up anything directly waiting upon the host.  this
1615          * will be requests for character device operations, and also for
1616          * ioctls to queued block devices.
1617          */
1618         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1619                                           __func__));
1620
1621         spin_lock_irqsave(shost->host_lock, flags);
1622         if (scsi_host_set_state(shost, SHOST_RUNNING))
1623                 if (scsi_host_set_state(shost, SHOST_CANCEL))
1624                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1625         spin_unlock_irqrestore(shost->host_lock, flags);
1626
1627         wake_up(&shost->host_wait);
1628
1629         /*
1630          * finally we need to re-initiate requests that may be pending.  we will
1631          * have had everything blocked while error handling is taking place, and
1632          * now that error recovery is done, we will need to ensure that these
1633          * requests are started.
1634          */
1635         scsi_run_host_queues(shost);
1636 }
1637
1638 /**
1639  * scsi_eh_ready_devs - check device ready state and recover if not.
1640  * @shost:      host to be recovered.
1641  * @work_q:     &list_head for pending commands.
1642  * @done_q:     &list_head for processed commands.
1643  */
1644 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1645                         struct list_head *work_q,
1646                         struct list_head *done_q)
1647 {
1648         if (!scsi_eh_stu(shost, work_q, done_q))
1649                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1650                         if (!scsi_eh_target_reset(shost, work_q, done_q))
1651                                 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1652                                         if (!scsi_eh_host_reset(work_q, done_q))
1653                                                 scsi_eh_offline_sdevs(work_q,
1654                                                                       done_q);
1655 }
1656 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1657
1658 /**
1659  * scsi_eh_flush_done_q - finish processed commands or retry them.
1660  * @done_q:     list_head of processed commands.
1661  */
1662 void scsi_eh_flush_done_q(struct list_head *done_q)
1663 {
1664         struct scsi_cmnd *scmd, *next;
1665
1666         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1667                 list_del_init(&scmd->eh_entry);
1668                 if (scsi_device_online(scmd->device) &&
1669                     !scsi_noretry_cmd(scmd) &&
1670                     (++scmd->retries <= scmd->allowed)) {
1671                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1672                                                           " retry cmd: %p\n",
1673                                                           current->comm,
1674                                                           scmd));
1675                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1676                 } else {
1677                         /*
1678                          * If just we got sense for the device (called
1679                          * scsi_eh_get_sense), scmd->result is already
1680                          * set, do not set DRIVER_TIMEOUT.
1681                          */
1682                         if (!scmd->result)
1683                                 scmd->result |= (DRIVER_TIMEOUT << 24);
1684                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1685                                                         " cmd: %p\n",
1686                                                         current->comm, scmd));
1687                         scsi_finish_command(scmd);
1688                 }
1689         }
1690 }
1691 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1692
1693 /**
1694  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1695  * @shost:      Host to unjam.
1696  *
1697  * Notes:
1698  *    When we come in here, we *know* that all commands on the bus have
1699  *    either completed, failed or timed out.  we also know that no further
1700  *    commands are being sent to the host, so things are relatively quiet
1701  *    and we have freedom to fiddle with things as we wish.
1702  *
1703  *    This is only the *default* implementation.  it is possible for
1704  *    individual drivers to supply their own version of this function, and
1705  *    if the maintainer wishes to do this, it is strongly suggested that
1706  *    this function be taken as a template and modified.  this function
1707  *    was designed to correctly handle problems for about 95% of the
1708  *    different cases out there, and it should always provide at least a
1709  *    reasonable amount of error recovery.
1710  *
1711  *    Any command marked 'failed' or 'timeout' must eventually have
1712  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1713  *    here, so when we restart the host after we return it should have an
1714  *    empty queue.
1715  */
1716 static void scsi_unjam_host(struct Scsi_Host *shost)
1717 {
1718         unsigned long flags;
1719         LIST_HEAD(eh_work_q);
1720         LIST_HEAD(eh_done_q);
1721
1722         spin_lock_irqsave(shost->host_lock, flags);
1723         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1724         spin_unlock_irqrestore(shost->host_lock, flags);
1725
1726         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1727
1728         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1729                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1730                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1731
1732         scsi_eh_flush_done_q(&eh_done_q);
1733 }
1734
1735 /**
1736  * scsi_error_handler - SCSI error handler thread
1737  * @data:       Host for which we are running.
1738  *
1739  * Notes:
1740  *    This is the main error handling loop.  This is run as a kernel thread
1741  *    for every SCSI host and handles all error handling activity.
1742  */
1743 int scsi_error_handler(void *data)
1744 {
1745         struct Scsi_Host *shost = data;
1746
1747         /*
1748          * We use TASK_INTERRUPTIBLE so that the thread is not
1749          * counted against the load average as a running process.
1750          * We never actually get interrupted because kthread_run
1751          * disables signal delivery for the created thread.
1752          */
1753         set_current_state(TASK_INTERRUPTIBLE);
1754         while (!kthread_should_stop()) {
1755                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1756                     shost->host_failed != shost->host_busy) {
1757                         SCSI_LOG_ERROR_RECOVERY(1,
1758                                 printk("Error handler scsi_eh_%d sleeping\n",
1759                                         shost->host_no));
1760                         schedule();
1761                         set_current_state(TASK_INTERRUPTIBLE);
1762                         continue;
1763                 }
1764
1765                 __set_current_state(TASK_RUNNING);
1766                 SCSI_LOG_ERROR_RECOVERY(1,
1767                         printk("Error handler scsi_eh_%d waking up\n",
1768                                 shost->host_no));
1769
1770                 /*
1771                  * We have a host that is failing for some reason.  Figure out
1772                  * what we need to do to get it up and online again (if we can).
1773                  * If we fail, we end up taking the thing offline.
1774                  */
1775                 if (scsi_autopm_get_host(shost) != 0) {
1776                         SCSI_LOG_ERROR_RECOVERY(1,
1777                                 printk(KERN_ERR "Error handler scsi_eh_%d "
1778                                                 "unable to autoresume\n",
1779                                                 shost->host_no));
1780                         continue;
1781                 }
1782
1783                 if (shost->transportt->eh_strategy_handler)
1784                         shost->transportt->eh_strategy_handler(shost);
1785                 else
1786                         scsi_unjam_host(shost);
1787
1788                 /*
1789                  * Note - if the above fails completely, the action is to take
1790                  * individual devices offline and flush the queue of any
1791                  * outstanding requests that may have been pending.  When we
1792                  * restart, we restart any I/O to any other devices on the bus
1793                  * which are still online.
1794                  */
1795                 scsi_restart_operations(shost);
1796                 scsi_autopm_put_host(shost);
1797                 set_current_state(TASK_INTERRUPTIBLE);
1798         }
1799         __set_current_state(TASK_RUNNING);
1800
1801         SCSI_LOG_ERROR_RECOVERY(1,
1802                 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1803         shost->ehandler = NULL;
1804         return 0;
1805 }
1806
1807 /*
1808  * Function:    scsi_report_bus_reset()
1809  *
1810  * Purpose:     Utility function used by low-level drivers to report that
1811  *              they have observed a bus reset on the bus being handled.
1812  *
1813  * Arguments:   shost       - Host in question
1814  *              channel     - channel on which reset was observed.
1815  *
1816  * Returns:     Nothing
1817  *
1818  * Lock status: Host lock must be held.
1819  *
1820  * Notes:       This only needs to be called if the reset is one which
1821  *              originates from an unknown location.  Resets originated
1822  *              by the mid-level itself don't need to call this, but there
1823  *              should be no harm.
1824  *
1825  *              The main purpose of this is to make sure that a CHECK_CONDITION
1826  *              is properly treated.
1827  */
1828 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1829 {
1830         struct scsi_device *sdev;
1831
1832         __shost_for_each_device(sdev, shost) {
1833                 if (channel == sdev_channel(sdev))
1834                         __scsi_report_device_reset(sdev, NULL);
1835         }
1836 }
1837 EXPORT_SYMBOL(scsi_report_bus_reset);
1838
1839 /*
1840  * Function:    scsi_report_device_reset()
1841  *
1842  * Purpose:     Utility function used by low-level drivers to report that
1843  *              they have observed a device reset on the device being handled.
1844  *
1845  * Arguments:   shost       - Host in question
1846  *              channel     - channel on which reset was observed
1847  *              target      - target on which reset was observed
1848  *
1849  * Returns:     Nothing
1850  *
1851  * Lock status: Host lock must be held
1852  *
1853  * Notes:       This only needs to be called if the reset is one which
1854  *              originates from an unknown location.  Resets originated
1855  *              by the mid-level itself don't need to call this, but there
1856  *              should be no harm.
1857  *
1858  *              The main purpose of this is to make sure that a CHECK_CONDITION
1859  *              is properly treated.
1860  */
1861 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1862 {
1863         struct scsi_device *sdev;
1864
1865         __shost_for_each_device(sdev, shost) {
1866                 if (channel == sdev_channel(sdev) &&
1867                     target == sdev_id(sdev))
1868                         __scsi_report_device_reset(sdev, NULL);
1869         }
1870 }
1871 EXPORT_SYMBOL(scsi_report_device_reset);
1872
1873 static void
1874 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1875 {
1876 }
1877
1878 /*
1879  * Function:    scsi_reset_provider
1880  *
1881  * Purpose:     Send requested reset to a bus or device at any phase.
1882  *
1883  * Arguments:   device  - device to send reset to
1884  *              flag - reset type (see scsi.h)
1885  *
1886  * Returns:     SUCCESS/FAILURE.
1887  *
1888  * Notes:       This is used by the SCSI Generic driver to provide
1889  *              Bus/Device reset capability.
1890  */
1891 int
1892 scsi_reset_provider(struct scsi_device *dev, int flag)
1893 {
1894         struct scsi_cmnd *scmd;
1895         struct Scsi_Host *shost = dev->host;
1896         struct request req;
1897         unsigned long flags;
1898         int rtn;
1899
1900         if (scsi_autopm_get_host(shost) < 0)
1901                 return FAILED;
1902
1903         scmd = scsi_get_command(dev, GFP_KERNEL);
1904         blk_rq_init(NULL, &req);
1905         scmd->request = &req;
1906
1907         scmd->cmnd = req.cmd;
1908
1909         scmd->scsi_done         = scsi_reset_provider_done_command;
1910         memset(&scmd->sdb, 0, sizeof(scmd->sdb));
1911
1912         scmd->cmd_len                   = 0;
1913
1914         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
1915
1916         spin_lock_irqsave(shost->host_lock, flags);
1917         shost->tmf_in_progress = 1;
1918         spin_unlock_irqrestore(shost->host_lock, flags);
1919
1920         switch (flag) {
1921         case SCSI_TRY_RESET_DEVICE:
1922                 rtn = scsi_try_bus_device_reset(scmd);
1923                 if (rtn == SUCCESS)
1924                         break;
1925                 /* FALLTHROUGH */
1926         case SCSI_TRY_RESET_TARGET:
1927                 rtn = scsi_try_target_reset(scmd);
1928                 if (rtn == SUCCESS)
1929                         break;
1930                 /* FALLTHROUGH */
1931         case SCSI_TRY_RESET_BUS:
1932                 rtn = scsi_try_bus_reset(scmd);
1933                 if (rtn == SUCCESS)
1934                         break;
1935                 /* FALLTHROUGH */
1936         case SCSI_TRY_RESET_HOST:
1937                 rtn = scsi_try_host_reset(scmd);
1938                 break;
1939         default:
1940                 rtn = FAILED;
1941         }
1942
1943         spin_lock_irqsave(shost->host_lock, flags);
1944         shost->tmf_in_progress = 0;
1945         spin_unlock_irqrestore(shost->host_lock, flags);
1946
1947         /*
1948          * be sure to wake up anyone who was sleeping or had their queue
1949          * suspended while we performed the TMF.
1950          */
1951         SCSI_LOG_ERROR_RECOVERY(3,
1952                 printk("%s: waking up host to restart after TMF\n",
1953                 __func__));
1954
1955         wake_up(&shost->host_wait);
1956
1957         scsi_run_host_queues(shost);
1958
1959         scsi_next_command(scmd);
1960         scsi_autopm_put_host(shost);
1961         return rtn;
1962 }
1963 EXPORT_SYMBOL(scsi_reset_provider);
1964
1965 /**
1966  * scsi_normalize_sense - normalize main elements from either fixed or
1967  *                      descriptor sense data format into a common format.
1968  *
1969  * @sense_buffer:       byte array containing sense data returned by device
1970  * @sb_len:             number of valid bytes in sense_buffer
1971  * @sshdr:              pointer to instance of structure that common
1972  *                      elements are written to.
1973  *
1974  * Notes:
1975  *      The "main elements" from sense data are: response_code, sense_key,
1976  *      asc, ascq and additional_length (only for descriptor format).
1977  *
1978  *      Typically this function can be called after a device has
1979  *      responded to a SCSI command with the CHECK_CONDITION status.
1980  *
1981  * Return value:
1982  *      1 if valid sense data information found, else 0;
1983  */
1984 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1985                          struct scsi_sense_hdr *sshdr)
1986 {
1987         if (!sense_buffer || !sb_len)
1988                 return 0;
1989
1990         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1991
1992         sshdr->response_code = (sense_buffer[0] & 0x7f);
1993
1994         if (!scsi_sense_valid(sshdr))
1995                 return 0;
1996
1997         if (sshdr->response_code >= 0x72) {
1998                 /*
1999                  * descriptor format
2000                  */
2001                 if (sb_len > 1)
2002                         sshdr->sense_key = (sense_buffer[1] & 0xf);
2003                 if (sb_len > 2)
2004                         sshdr->asc = sense_buffer[2];
2005                 if (sb_len > 3)
2006                         sshdr->ascq = sense_buffer[3];
2007                 if (sb_len > 7)
2008                         sshdr->additional_length = sense_buffer[7];
2009         } else {
2010                 /*
2011                  * fixed format
2012                  */
2013                 if (sb_len > 2)
2014                         sshdr->sense_key = (sense_buffer[2] & 0xf);
2015                 if (sb_len > 7) {
2016                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2017                                          sb_len : (sense_buffer[7] + 8);
2018                         if (sb_len > 12)
2019                                 sshdr->asc = sense_buffer[12];
2020                         if (sb_len > 13)
2021                                 sshdr->ascq = sense_buffer[13];
2022                 }
2023         }
2024
2025         return 1;
2026 }
2027 EXPORT_SYMBOL(scsi_normalize_sense);
2028
2029 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2030                                  struct scsi_sense_hdr *sshdr)
2031 {
2032         return scsi_normalize_sense(cmd->sense_buffer,
2033                         SCSI_SENSE_BUFFERSIZE, sshdr);
2034 }
2035 EXPORT_SYMBOL(scsi_command_normalize_sense);
2036
2037 /**
2038  * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
2039  * @sense_buffer:       byte array of descriptor format sense data
2040  * @sb_len:             number of valid bytes in sense_buffer
2041  * @desc_type:          value of descriptor type to find
2042  *                      (e.g. 0 -> information)
2043  *
2044  * Notes:
2045  *      only valid when sense data is in descriptor format
2046  *
2047  * Return value:
2048  *      pointer to start of (first) descriptor if found else NULL
2049  */
2050 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2051                                 int desc_type)
2052 {
2053         int add_sen_len, add_len, desc_len, k;
2054         const u8 * descp;
2055
2056         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2057                 return NULL;
2058         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2059                 return NULL;
2060         add_sen_len = (add_sen_len < (sb_len - 8)) ?
2061                         add_sen_len : (sb_len - 8);
2062         descp = &sense_buffer[8];
2063         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2064                 descp += desc_len;
2065                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2066                 desc_len = add_len + 2;
2067                 if (descp[0] == desc_type)
2068                         return descp;
2069                 if (add_len < 0) // short descriptor ??
2070                         break;
2071         }
2072         return NULL;
2073 }
2074 EXPORT_SYMBOL(scsi_sense_desc_find);
2075
2076 /**
2077  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
2078  * @sense_buffer:       byte array of sense data
2079  * @sb_len:             number of valid bytes in sense_buffer
2080  * @info_out:           pointer to 64 integer where 8 or 4 byte information
2081  *                      field will be placed if found.
2082  *
2083  * Return value:
2084  *      1 if information field found, 0 if not found.
2085  */
2086 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2087                             u64 * info_out)
2088 {
2089         int j;
2090         const u8 * ucp;
2091         u64 ull;
2092
2093         if (sb_len < 7)
2094                 return 0;
2095         switch (sense_buffer[0] & 0x7f) {
2096         case 0x70:
2097         case 0x71:
2098                 if (sense_buffer[0] & 0x80) {
2099                         *info_out = (sense_buffer[3] << 24) +
2100                                     (sense_buffer[4] << 16) +
2101                                     (sense_buffer[5] << 8) + sense_buffer[6];
2102                         return 1;
2103                 } else
2104                         return 0;
2105         case 0x72:
2106         case 0x73:
2107                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2108                                            0 /* info desc */);
2109                 if (ucp && (0xa == ucp[1])) {
2110                         ull = 0;
2111                         for (j = 0; j < 8; ++j) {
2112                                 if (j > 0)
2113                                         ull <<= 8;
2114                                 ull |= ucp[4 + j];
2115                         }
2116                         *info_out = ull;
2117                         return 1;
2118                 } else
2119                         return 0;
2120         default:
2121                 return 0;
2122         }
2123 }
2124 EXPORT_SYMBOL(scsi_get_sense_info_fld);
2125
2126 /**
2127  * scsi_build_sense_buffer - build sense data in a buffer
2128  * @desc:       Sense format (non zero == descriptor format,
2129  *              0 == fixed format)
2130  * @buf:        Where to build sense data
2131  * @key:        Sense key
2132  * @asc:        Additional sense code
2133  * @ascq:       Additional sense code qualifier
2134  *
2135  **/
2136 void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2137 {
2138         if (desc) {
2139                 buf[0] = 0x72;  /* descriptor, current */
2140                 buf[1] = key;
2141                 buf[2] = asc;
2142                 buf[3] = ascq;
2143                 buf[7] = 0;
2144         } else {
2145                 buf[0] = 0x70;  /* fixed, current */
2146                 buf[2] = key;
2147                 buf[7] = 0xa;
2148                 buf[12] = asc;
2149                 buf[13] = ascq;
2150         }
2151 }
2152 EXPORT_SYMBOL(scsi_build_sense_buffer);